ChatGPT and Claude cite Grokipedia as source for information
Tests by The Guardian found that ChatGPT and Claude increasingly reference Grokipedia, Elon Musk's AI-generated encyclopedia, as a source for information. Experts warn this creates a "fact-washing" effect where ideologically biased or unverified claims gain false credibility when cited by mainstream AI systems.
What happened
Tests conducted by The Guardian showed that ChatGPT increasingly references Grokipedia, an online encyclopedia created and edited by Elon Musk's xAI using AI, as a source for information.
In tests covering topics including Iranian politics, historical figures, and the January 6 Capitol attack, ChatGPT cited Grokipedia. For niche topics, ChatGPT cited claims from Grokipedia including that an Iranian mobile operator had direct connections to the office of the Supreme Leader, a claim not found on Wikipedia, and repeated refuted details about Holocaust denier David Irving.
Claude, Anthropic's AI model, also cites Grokipedia as a source.
Reported impact
Disinformation researcher Nina Jankowicz warned of a "fact-washing" effect: information that appears ideologically biased on Grokipedia gains false credibility when cited by ChatGPT, leading users to believe ChatGPT has vetted the information when it has not.
Unlike Wikipedia, Grokipedia has no human editors to review content and has been criticized for presenting right-wing populist narratives uncritically, including claims about the Capitol attack and same-sex marriage.
Index analysis
Experts warn of a dangerous AI feedback loop where AI-generated content from Grokipedia is cited by other AI systems, potentially amplifying misinformation.
The integration occurs largely invisibly; direct queries about sensitive topics like January 6 trigger security filters that prevent Grokipedia from appearing as a source, but on niche topics the citations occur without such filtering.
Governance controls
ChatGPT's safety filters prevent Grokipedia from appearing as a source for direct queries about sensitive political topics, but these controls do not extend to niche or less-monitored subjects.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.